Agenda

Round-table update on analysis for the Predictive processing data release manuscript

We plan to start with a discussion around some analysis that David Wyrick started.

In addition, please submit slides material prior to the meeting at the discussion link at the bottom of this meeting page.

Meeting Recording

.

Meeting Notes

OpenScope Renewal And Community Support: Jerome described the planned NIH U24 renewal that will transition OpenScope into the Global Brain Observatory, while Stefan asked whether support from Allen Institute staff would help and Jerome encouraged all interested contributors to sign the support letter.

Renewal Proposal: Jerome said the NIH U24 renewal will focus on supporting the community around the observatory. Existing foundation support is expected to fund experimental work next year, and the project is planned to be renamed the Global Brain Observatory.

Community Resources: The requested NIH support would help organize workshops, provide data fellows, and offer cloud computing resources for community analysis. Jerome said coordination of these resources has been difficult.

Support Letter: Jerome asked people at all career levels who support the community-project goals to read and sign the letter submitted with the application. In response to Stefan’s question, Jerome said internal Allen Institute support could also help and encouraged Stefan to contribute if comfortable.

Neuropixels Responsiveness Framework: David presented context-specific definitions and an interactive viewer for quantifying mismatch responsiveness in Neuropixels recordings, and Jerome, Stefan, and David agreed that the criteria should undergo documented review before being merged into the manuscript and data release.

Context-Specific Comparisons: David explained that responsiveness must be defined separately for each stimulus context: standard oddball compares a deviant with the preceding expected stimulus; sequence mismatch compares the substituted element with the corresponding element in the prior sequence; duration mismatch compares post-delay stimuli and may also require a separate delay-period test; and sensorimotor mismatch compares activity immediately before visual-flow decoupling because the flow is continuous.

Data Availability: David reported that only one or two Neuropixels animals had sufficient running during sensorimotor mismatch events to quantify responses, compared with roughly ten mesoscope animals. He cautioned that example figures based on one running animal are not representative of the full Neuropixels dataset.

Interactive Viewer: David added responsiveness filtering to the Neuropixels event-response viewer, which allows users to select mismatch blocks, events, anatomical areas, quality-control status, and unit classes. In one example, 127 of 470 possible cells were identified as responsive to a 45-degree orientation substitution.

Review And Merge: Jerome asked whether the definitions were documented and ready for manuscript use. David said they are described in an open pull request and markdown summary, and that he had incorporated them into a manuscript draft but had not merged them because the methodological decisions still need review. Jerome proposed preparing a clean pull request and manuscript version for interested reviewers; David estimated that this could be ready within about a week, and Stefan said formulas and a complete written description would make review easier.

Mesoscope Follow-Up: Jerome said the limited Neuropixels sensorimotor data should not end the analysis and proposed repeating it with mesoscope mice, which contain more qualifying running events. The team also noted that unusual session-level data constraints need to be documented so future users do not have to rediscover them.

WaveMAP Data Navigation: marianna presented an area-specific WaveMAP organization of extracellular waveforms intended to help users navigate unit diversity, anatomical information, and waveform metadata without treating every cluster as a distinct biological cell type.

Area-Specific Clustering: marianna described normalizing each quality-control-passing unit’s mean extracellular waveform and applying WaveMAP independently within major anatomical groups, including motor, prefrontal, visual, striatal, hippocampal, and thalamic regions. Classes are local to each anatomical area, so identically numbered classes across areas are not intended to represent the same population.

Clustering Resolution: marianna said the team tested multiple clustering resolutions and selected a conservative representation to reduce over-splitting. The parameters remain adjustable, and the team had also considered excluding very small clusters, such as clusters containing fewer than 20 waveforms.

Interactive Exploration: The interactive explorer allows users to select an anatomical group, inspect the WaveMAP embedding, hover for metadata, click individual units, view extracellular waveforms against class waveforms, and examine cortical-layer profiles where available. This links population-level waveform organization back to individual units.

Depth Analysis: Jerome asked whether waveform clusters showed depth gradients even though depth was not used as a clustering input. marianna confirmed that the current analysis uses waveform shape only and said the depth distribution would need to be examined separately; Yuhang is working with Sarah on depth analysis.

Waveform Classes And Cell-Type Annotations: marianna compared putative fast-spiking/PV-like units with optotagged SST units in WaveMAP space, while Jerome, Lucas, Alexander, and Stefan discussed the limits of waveform-based cell-type inference and possible extensions using genetic or supervised information.

Population Definitions: marianna defined putative PV-like units using narrow waveforms and high firing rates, while SST identity came from optotagging. Putative PV-like units occupied a relatively distinct region in cortical WaveMAP spaces, whereas optotagged SST units were more broadly distributed and partly overlapped with the narrow-spiking population.

PV Tagging Clarification: Lucas asked whether PV cells were also optotagged in this dataset. marianna and Jerome clarified that the presented dataset includes optotagged SST cells but not optotagged PV cells; the PV-like group is therefore a proxy based on extracellular waveform and firing-rate criteria.

Biological Interpretation: Stefan asked whether WaveMAP clusters were expected to correspond to genetic cell types or instead to recording geometry and electrode approach. Sarah clarified that the current analysis uses only the waveform at the peak channel and does not include spatial spread across neighboring channels; she said spatial waveform features could be explored, but may reflect electrode geometry and propagation effects rather than reliably resolving cell subtypes.

Supervised Extension: Stefan proposed using tagged populations to construct supervised discriminative axes, such as linear discriminant analysis, if enough labeled examples are available. Sarah acknowledged this as a possible direction, while the discussion emphasized that optotagging or other biological annotations would remain important for validating cell-type interpretations.

Future Multimodal Linkage: Lucas suggested linking WaveMAP classes with planned genetic information to extrapolate annotations into sessions without direct tagging. Jerome said a future presentation would cover multiplex FISH data from a subset of imaging mice and noted that WaveMAP could then be applied to compare waveform and genetic-class distributions.

Waveform Composition Across Contexts: marianna examined whether local waveform classes were represented similarly across the four predictive-processing contexts, and Jerome redirected the interpretation toward context-dependent engagement and firing rather than simple unit proportions.

Context Comparison: marianna compared the proportion of units in each area-specific WaveMAP class across standard oddball, sequence mismatch, duration mismatch, and sensorimotor mismatch contexts. Proportions were normalized within recording sessions and then averaged across sessions.

Observed Distributions: The analysis did not show one context that consistently stood out across all waveform classes, although some individual classes appeared unevenly represented or absent in particular contexts, such as a thalamic class.

Scientific Interpretation: Jerome questioned whether waveform-defined units should be expected to be equally represented across contexts because waveform is primarily a unit property. He suggested that a more informative analysis would measure how strongly those units engage in each context through their firing responses, rather than relying only on class proportions; marianna agreed.

Manuscript Integration: Jerome said the WaveMAP figures can educate readers about waveform diversity, anatomical organization, and available cell annotations without needing to prove that each class has a specific function. He asked marianna to submit the figures and code as a pull request within about a week so they can be integrated with Roberto’s optotagging work into manuscript panels.

Manuscript And Meeting Follow-Up: Jerome planned to consolidate the recent analyses into a cohesive manuscript presentation for the following week, followed by a multiplex FISH discussion two weeks later, while assigning near-term review and integration work to David and marianna.

Next Presentation: Jerome said he would spend the next week integrating the recent analytical snapshots into a coherent story and present that synthesis at the next meeting so participants could identify issues and iterate on the draft.

Upcoming FISH Discussion: Jerome said the meeting after that would include a presentation and discussion of multiplex FISH data from imaging mice, providing genetic-class information relevant to the WaveMAP work.

Immediate Deliverables: David is expected to prepare a clean, reviewable pull request describing the responsiveness definitions, and marianna is expected to share the WaveMAP analysis code and figures for possible manuscript integration.

💬 Join the discussion for this page on GitHub